Precision Health 101—Understanding the Keys to Value

In May, I participated in a Tweet chat called “Precision Health 101 – Understanding the Keys to Value.” While Twitter provides a useful platform for engagement on broad concepts, its brevity necessarily limits the depth of discussion. I thought it might be useful to expand on some of the key concepts that were discussed during the chat.

For example, one of the questions was, “What should an organization be doing to prepare for and participate in precision health?” My response was that it’s not really about decision support, it’s about accessing information and navigating the healthcare ecosystem.

Often, when people think about precision health, they think about the development of technologies that can help tell doctors to know what to do in complex situations. However, there are a number of challenges associated with the development of data-driven decision support tools. One is that every algorithm has strengths and weaknesses, and will get something wrong once in a while. For example, every algorithm will sometimes produce a false positive by recommending something that, in the physician’s opinion, isn’t the best next step.

Integrating precision health technology into the clinical setting isn’t really about telling doctors what to do – it’s about developing algorithms to give doctors all of the information they need when they need it.

To support that concept, health care organizations are investing in systems that can connect data from across their organization. Technologies are readily available today that can recognize a concept across different data sets. For instance, within an organization, a glucose measurement might be coded one way in one system (for example “glucose82948”), and a different way in another system (”67777W”). Cutting-edge analytical technologies can now mine the various silos within an organization and consolidate that data in the EMR, in ways that are useful to clinicians. And yes, these are both real codes for glucose measurements.

Precision Health

During the Tweet chat, another question was, “How will data and analytics impact precision health? I replied “Without data, there is no precision health. What makes it precise is having the right information for each care context.”

Precision medicine is really all about knowing the details of a patient’s health history well enough to be able to make the optimal decisions for that individual patient. But there are fundamental challenges that can deny clinicians the full picture – for example, what is the patient’s complete and accurate problem list? What are the labs and diagnostics showing? What are the changes to those diagnostic markers over time?

In many hospitals and practices, there are islands of information that aren’t connected. Even in the rare case that clinical information systems talk to each other, if data isn’t coded consistently across an organization, it can be challenging to pull all of the relevant information together. There are companies and technologies that are evolving very rapidly to fill that niche – Apixio and Lumiata, for example, offer analytics suites that provide natural language processing, pattern recognition, and machine learning to help health care providers build comprehensive views of their aggregated EMR data.

Not long ago I worked on a study aggregating data from multiple healthcare sources in the context of a patient’s problem list. The study found that up to 63 percent of the time, aggregated problem lists are missing key clinical information, such as the history of a heart attack. The study also found that up to 50 percent of the time, conditions such as heart failure were incorrectly flagged as major issues on the problem list. In other words, many EMRs contain false positives and false negatives, which means that the very first priority for precision medicine is to create a full, accurate and reliable view of the patient across all relevant sites of care.

Machine Learning

It's a great machine learning problem. One way to fix errors in the EMR is to use machine learning and advanced analytics to understand what the doctor actually wrote to flag errors and inconsistencies in structured EMR data. The care system knows what it’s doing; the problem is that the structured data in the EMR is influenced by billing procedures and is not specifically designed to accurately capture the patient’s clinical situation. AI, machine learning, and text analytics are great tools for figuring out what’s really going on with each patient.

Where is the intersection between big data and precision healthcare for your organization? What questions do you have?